[FoRK] Super-Intelligent Humans Are Coming

J. Andrew Rogers andrew at jarbox.org
Mon Oct 27 10:35:43 PDT 2014

> On Oct 26, 2014, at 10:32 PM, Stephen D. Williams <sdw at lig.net> wrote:
>> Tangentially, we know almost nothing about mirror neurons, and certainly not what you are implying. Most things ascribed to them in the popular media is pseudo-science. Like EQ, it enables a pleasing Gladwell-ian narrative with a truthiness that real science fails to deliver.
> Another example of specialized subsystems, one that we do seem to have very specific proof for, are grid cells:
> http://www.livescience.com/38772-human-gps-neurons-found.html <http://www.livescience.com/38772-human-gps-neurons-found.html>

Yes, I know, but it is not relevant here. Something like grid cells are theoretically predicted and required (see below).

> In any case, most emotional queue processing and reaction seems to be subconscious and automatic.  

It is only semi-autonomous, like breathing, at least in higher animals. There is considerable ability to do it consciously if useful or desired. Empirically, it is computationally trivial and shows up in animals with much simpler brains. Again, it doesn’t have much to do with intelligence and doesn’t require any special neural machinery to do well. 

> I'm fairly surprised that we have something as specific as grid cells in a particular location for such an important purpose.  It hints at more specialization.  Clearly we have one or more generalization layers on top of these things too.

Why are grid cells surprising? Conventional neural structures compute over graph-like relationships. Spatial reasoning is not expressible over a graph. The existence of an alternative computing structures suitable for spatial reasoning is not surprising given that animals have excellent spatial reasoning. Specialized emotional processing does not follow from the existence of grid cells since emotional processing is not computationally different or difficult. 

As an aside, the ubiquity with which computer science tries to cram spatial computing into graph-like idioms goes a long way toward explaining why most spatial software systems have dreadful scalability. You *can* efficiently cram graph computing into spatial structures but neural biology makes this vastly less efficient than it is on silicon, hence the specialization. 

More information about the FoRK mailing list